Anthropic — AI supply-chain exposure

Anthropic is a frontier model developer built around reliability and safety — its Claude models aimed at high-stakes enterprise and developer use, where predictable, well-behaved output matters more than novelty. Like its peers it trains large models and then serves them, sitting on the model layer as both a producer of frontier systems and a heavy consumer of the inference compute that runs them.

Its structural hook is the same token economics that govern any serving business — cost per token, and the agent workflows that stretch a single request into long chains of model calls — but weighted toward customers who pay for dependability over raw scale. That skew toward enterprise and coding workloads shapes what it optimises for and who it sells to. The model places it alongside the other frontier labs because it turns compute into finished intelligence, and its growth is gated by the same access to chips and inference capacity that constrains every lab that does not make its own silicon.